WealthDesk -- In-Class Build

Agentic AI Engineering, Batch 1 (June 2026)

WealthDesk is the AI banking assistant for Bharat National Bank (BNB) that you build session by session across the first 16 sessions of the course. By Session 16 it is a multi-agent system with a Streamlit UI, ChromaDB RAG, SQLite tool calls, MCP integration, compliance filtering, LangSmith observability, and a Docker deployment.

---

Getting Started

Step 1 -- Fork this repo (click Fork at the top right on GitHub). This gives you your own copy where you can save your work.

Step 2 -- Clone your fork (replace YOUR_USERNAME with your GitHub username):

git clone https://github.com/YOUR_USERNAME/wealthdesk.git
cd wealthdesk

Step 3 -- Add the instructor repo as upstream so you can pull new session files before each class:

git remote add upstream https://github.com/ketanvj/wealthdesk.git

Before each session, pull the latest starter files:

git pull upstream main

After working in class, save your code to your fork:

git add .
git commit -m "Session X work"
git push

---

What You Need Before Session 1

1. Python 3.11 or later
2. A Groq API key (free tier): https://console.groq.com/keys
3. VS Code with the Python extension

Setup (run once, before Session 1):

# Windows
pip install -r requirements.txt
copy .env.example .env
# Edit .env with your actual Groq API key

# Mac/Linux
pip install -r requirements.txt
cp .env.example .env
# Edit .env with your actual Groq API key

Seed the database and vector store (run once, before Session 1):

python data/seed.py
python data/ingest.py

Verify your setup:

python test_setup.py

All checks should show OK. Take a screenshot and post it in the WhatsApp group.

---

Folder Structure

wealthdesk/
  data/
    documents/          5 BNB policy documents (ChromaDB source)
    seed.py             Creates and seeds data/bnb_data.db
    ingest.py           Ingests documents into data/vectorstore/
    bnb_data.db         Generated -- do not commit (in .gitignore)
    vectorstore/        Generated -- do not commit (in .gitignore)

  s01/                  Session 1: Basic Conversational Agent (US-01)
    starter/main.py     Your starting point -- fill in the TODOs
    solution/main.py    Reference solution
    tests/test_s01.py   Unit tests (run with: pytest s01/tests/ -v)
    instructor_notes.md Session plan, timings, common issues

  s02/ ... s16/         Added as the course progresses

  requirements.txt      All packages for all 17 sessions (install once)
  .env.example          Template for your .env file
  .gitignore            Excludes .env, database, and vector store from git

---

Session Map

Session	Story	What You Build
Pre-S1	US-00	Data layer: SQLite + ChromaDB setup (seed.py, ingest.py)
S1	US-01	Terminal chatbot with single-turn Groq + LangGraph
S2	US-02	Multi-turn memory with SQLite checkpointer
S3	US-07	Query routing: SIMPLE vs COMPLEX, RM escalation
S4	US-03	ChromaDB RAG: policy document retrieval
S5	US-04	SQLite tool calls: live rate and branch lookup
S6	US-05	Baseline evaluation: 40-question golden dataset, LLM-as-judge
S7	US-06 pt.1	MCP server setup and tool wrapping
S8	US-06 pt.2	Agent calls tools via MCP protocol
S9	US-08+10	Compliance filter (SEBI/DPDP) + LangSmith tracing
S10	US-11 pt.1 + US-09	Multi-agent: Supervisor, Documents Agent, Rates Agent
S11	--	Industry guest session (no build)
S12	US-11 pt.2	Multi-agent system complete: perf gate + Streamlit skeleton
S13	US-12+16	Streamlit UI + Human-in-the-Loop approval
S14	US-14	Security and guardrails (OWASP LLM Top 10 + DPDP Act)
S15	US-13	Dockerfile + Docker run + cloud deployment
S16	US-15+17	Advanced eval + data flywheel + prompt versioning

---

Running the Tests

# From the wealthdesk/ directory -- run one session at a time:
pytest s01/tests/ -v
pytest s02/tests/ -v
pytest s03/tests/ -v

# Tests do not require a live Groq API key -- the LLM is mocked.

Note: always run one session's tests at a time. All sessions use a file named main.py, so running multiple sessions together (e.g. pytest s01/ s02/) can cause the wrong main to be patched. Per-session runs are always correct.

---

Important Rules

• Never commit your .env file. It contains your API key.
• Always run scripts from the wealthdesk/ directory, not from inside a subfolder.
• If you accidentally expose an API key on GitHub, rotate it immediately at the provider's console.
• When in doubt about a rate or policy, the database is the source of truth -- not the documents, not the system prompt.

---

Getting Help

Post in the Batch 1 WhatsApp group. Include: the error message, which script you were running, and which operating system you are on.